PegaWorld 2025: Harnessing the Power of Agentic AI: Introducing Pega Self Service Agent for Transformative Customer Engagement

PegaWorld 2025: Harnessing the Power of Agentic AI: Introducing the Pega Self-Service Agent

Thank you everyone for coming today and we are very grateful for your time. And we know there's lots of choice today. We're very grateful. Coming to hear how Pega is harnessing the power of Agentic AI for exceptional customer self-service. I'm Paul Conlin. I work in the go to market strategy at Pega and I'm joined by Jason Ouimette.

I lead our conversational AI products on the customer service team. If any of you attended the previous session, it's quite ironic because the products that I manage directly compete with each other. I manage the solutions that try and make the AI assisted agent technologies, which I'm not allowed to say AI assisted CSR technologies.

And now our new Agentic solutions, which when properly implemented, means you need significantly less representatives in your workforce. So as you can hear from my accent, I'm from Scotland. So I'm going to start off with a sport analogy. And many people in Scotland like football, sorry. Soccer for American friends.

And we're a proud nation and we want to compete at the highest level. We want to do well. I've used some artistic license here in saying that Scotland could win the World Cup. Chances of that are very, very slim. But for most of these big tournaments, we enter into them with many hopes and dreams of achieving great things.

What generally happens after a few weeks is glorious failure. Despite a few happy moments in victories, we normally fall short, and that's a pattern we see with many organizations when they've initiated self-service programs start off with great hopes and dreams of driving amazing customer experiences and reducing costs from calls into the contact center and making their agents much happier because they're no longer dealing with those mundane, simple calls.

But in reality, what happens is they fall short. The technology wasn't maybe there to drive the enterprise scale. Despite all the hard work and endeavor, it just doesn't live up to expectations. And that's generally what we see with a lot of self-service initiatives that are so out there in the market.

So start off with the worst of self-service, which usually comes in the voice channel. You ever tried voice recognition technology? No, they don't do Scottish accents. 11. Could you please repeat that? 1111. 1111. Could you please repeat that? So obviously I've experienced that many times, but hopefully not just me.

We've all had that experience where we engage with our voice self-service system. It's just an effective it's not serving the needs. You speak to the system, they eventually connect into a human agent who then have to repeat all the same things again. Similar experiences in very many bad chatbot deployments out there where it keeps asking you to rephrase the question, give me knowledge.

Information that's already in the website. And again, not servicing you. The problem being is that to deliver a really effective self-service application to now is complicated. The implementations are complicated. They're expensive, not to mention the ongoing cost to maintain these systems. And the result is, as we all have experienced, is very poor self-service experiences.

But that was the past. That ambition now is reached coming reality because technology is now catching up. Well, now the next generation of self-service, where the technology can actually meet the needs of customer service and representatives and move on through how the speech, the technology is meeting those recommendations.

So speech recognition has improved considerably over the last five years. Almost the point you can see here, it's almost flatlined. It's not perfect. My daughter's called Alyssa and many times my house. Alexa will answer when I'm asking her a question and she knows me. Speech recognition is very, very good, actually.

As you see, it's almost plateaued. But why are we not seeing that in An enterprise grade customer service. We've got a plethora of speech recognition devices in our lives, but we're not really seeing that in enterprise grade self-service. The reason behind that is the way in which we build these self-service applications.

The natural language processing, fundamentally, that's a very manual process. We have to build in the dialog and understand exactly how that human being is going to engage with the machine and build through the different complexities of conversation. The conversation isn't linear, so many times it goes off piste and the service breaks down.

And just to chime in, I mean, typically with these technologies, you're building them in a very must follow this structure manner, right? If the the customer on the other end tries to deviate at all outside of that tree. Good luck. Sorry, you must answer this question. If you don't answer this question, I can't help you.

And you know, it's one of those pieces that makes it really difficult, because it's clearly not a conversational All experience that you would typically have with another human. Yeah. And not only that, we have to train these models around every way in which that customer engages with machine. For example, generic customer service experience change of address.

We have to then train the model every different way a customer may ask change address. I'm moving home. Wouldn't necessarily the machine wouldn't know. That means you're changing address. So every single way in which that customer engage, we need to train these models to understand it. That as you can imagine, that's lots of datasets, lots of heavy lifting prior to Pega.

I worked in a selling and delivering speech recognition technologies and I have the war wounds to around us. DHL, the global logistics company. We rolled out a voice self-service application. We started off in the UK with English again, long project timelines, lots of design, testing, training and we go live? The system went live with accuracy at around 70%.

We had to do more training over time to get it more accurate. Not only have the project costs, the ongoing running costs of maintaining that system is very expensive. We then moved to France, did the same, but there was no reuse. We're building again. So despite the project being a success, the business case started weaken and weaken as we moved into lower revenue countries.

Eventually, the program was stopped because the business case did not stand up when we got to the smaller revenue countries for DHL. And that's where the revolution in technology is changing and the gigantic large language models, lots of that heavy lifting is taken away. No longer do we have to build in these very rigid compensation flows.

We give these agentic bots a goal. The objective for DHL, you are a customer service representative who needs to help this customer to find out whether delivery is And to make a delivery schedule a pickup and the bot can, then we don't have to train it because it's built in large language models. And the the, the implementation timelines are decreased dramatically.

And to bring that to life in terms of the old versus the new, we can walk through where projects that were months now and weeks, weeks become days. That massive training overhead is gone and it can now understand multi step conversations. If you imagine you want to book, for example, I want to book a hotel in Orlando, Florida next week.

Traditional way of building conversation flows would have to then take you through a sequential process. Okay which day are you going to for how many nights? Who's going with you? What type of room do you want? And gathering all that information. Think of that in a generic world. It's very much more proactive.

Active, the bot can be more and much more engaging. It can reply back based on your existing booking. I can see you're flying to a land on Friday, return on Sunday. Would you like to book your hotel based upon based upon your current loyalty points near Disney? And I assume your family are going with you very much, much more of a human like conversation, rather than that very robotic engagement you'd have with a traditional self-service application.

But as much as these experiences are much more richer and as much as powerful they are at Pega, we think there's one missing piece of the jigsaw. These bots, as intelligent as they are, they don't have that consistency. The predictability that Alan touched upon this morning around your business processes, they need to structure to understand, okay, how do I then fulfill that request? How would I then make a booking for that customer, the DHL example that I had, we could build a workflow underneath that and it would then know, I know the stages and the steps to find out where that parcel is rescheduled delivery, and do it on behalf so that all these bots are following the same predictable, consistent process.

And that's where we see the workflow as a fuel that powers these conversational agents to a perfect marriage of two great technologies within Agentic AI and workflow that we've always had in Pega. Pega always used to say in the strapline is we get work done and that's not changed. The workflow gets work done, but now with a much more conversational and natural experience for your customers.

So we move into you know, it's funny, you saw this slide today in the keynote, but a little bit different here because I like to refer to it as Center-out on steroids. And the reason for this is like if there's anything in the in the industry that can really articulate Center-out. It's this agentic solution.

So let's talk about what most of you have done. Some most of you have built some form of self- service across individual channels. You might have a self-service for your chat bots. You might have self-service for email, you might have a like, as Paul has mentioned, build out a self-service solution and a voice bot that's using natural language processing.

And if you're really advanced, you might have been able to combine all those into one flow, like there's a few solutions out there that will allow you to combine that. And so where you can do the same type of flow handling for voice, for chat, nothing but what it misses and what you'd always have to do something separate with is what do you do when it gets connected to the human? You're building out a different set of rules and workflows there, and it's a much more complex type. And so when we talk about Center-out on steroids, this solution is truly Center-out because you're building the The workflow for all of your channels, whether that be for self-service, for voice, digital messaging, email or you're using that same workflow for any reason, it escalates. Look, we have a great story that we'll show you around here about what we're going to do with Agentic, right? But we all know that there are going to be people that are going to refuse to interact with an AI solution.

There's going to be cases of it. Some people, younger generation, are going to be more accepting of it. There's going to be people that are like, I don't care. I want to talk to a human. The great thing about that is that same process that you built within Pega and that whole workflow application stack.

It works for all of the channels as well as for your CSR. So you're doing it one time and not having to worry about keeping those rules consistent in other pieces. And so when we talk about Center- out, it's not only Center-out with the channel and the workflow, it's the Center-out with all of the tools that your CSRs can interact with.

So when you talk about Knowledge Buddy and gaining the, you know, access to the content or the different types of systems that your Pega or Pega solution is connecting to, whether that be to other APIs or, God forbid, robotics. Process automation to a green screen like ARIN showed you today, right? This is this is all there.

And you know, one thing I want to mention with this too, is I was surprised, you know, Kerim did two things. Kudos to him. He is as brave as brave could be. That demo he showed you today where he called into Blueprint. That was real. That was live. That was our real product. But what he didn't tell you was, that is not just for Blueprint we have you will have at CA come out this year.

We have a voice bot solution that you will be able to leverage. So typically you saw that Pega would deliver a what we call a chatbot, right? Our for our digital messaging or email automation. But now we're leveraging these agentic solutions not only for those channels now enhancing those, but we're using those Agentic solutions for voice as well.

Um, you know, the the the key with this is that, uh, you're not having to rebuild all of your workflows when we do that adaptation. So when we go into how's it work? This is where I think Alan touched base on it. Kerim did as well. This is where we feel like we have a real advantage. A lot of the solutions that you see today that are talking about agentic really are just prompts.

They're they're building on prompts about these steps. And even in, in some of the technologies out there that are leveraging Agentic solutions and handling the communications, it still follows that tree mentality. But doing so with prompts. And it's really complicated. The advantage here is that we're using our workflow.

And then also the AI is thinking, right. It's using reasoning along with the rules and the elements that you've defined in your application stack of being able to workflow and get those details right. So you're getting predictable AI eye with AI solutions that truly are unpredictable. I'll give you a good example of it, right.

You know, um, if you want to know unpredictability, it's a really good experience. Go down. Try the customer experience today and in Innovation Hub, try it tomorrow. Try it today and then try it tomorrow. I guarantee you, if you pick the same persona, if you are unfamiliar, we have different voices you can interact with and different challenges, but you can pick the same exact one. Your experience will be different between both of them. It will not be exactly the same. It always changes. It's always different. And what we have delivered with this in terms of the solution is we've given you a solution that even though you have that slightly different variation, you have predictable results as to what you're going to get with that solution.

You have predictable results as to what you should expect in terms of the data you collected and workflow. And then on top of that, think about Paul's. Paul's scenario. Right. It's able to build in that human touch. So Paul mentioned, hey, we're going to Orlando, and now we want to book a reservation because we're going to, you know, we're going to leave on Sunday.

I'm going I want to be by Disney. Can you imagine now with an agentic solution, you could turn this into a customer satisfaction item to nurture them, or perhaps even an upsell scenario where the agentic solution goes and looks up the weather for Orlando. Because what if they're booking this week, right.

We can do a predictability. We go out and look at the weather and we go, oh, hey. Well, Paul, I wanted to let you know, I'm sorry. There's a weather warning on Sunday. Would you like, uh, or on Monday? Let's say they're checking on Monday. Would you like to check out early on Sunday? Or would you like us to extend your reservation to Tuesday? Because it's unlikely that you'll be able to fly out on Monday, right.

You can now do those capabilities. And quite honestly, that's also a leg up on on some of the human interactions. If you called into a travel agency today, do you think they're going to go look up the weather and tell you about any issues that potentially could arise. You're able to build these really compelling solutions that can improve the experience, so that people do want to interact with this technology, because you're giving them a much more compelling solution to come into place with it.

Um. And it's. The most powerful aspect of this, especially for Pega customers, is you're leveraging, leveraging your existing investment in Pega workflows. Now, what we spoke about there is agent agent calling in a workflow. These workflows can then pull in different agents and orchestrate the whole end to end process.

It's not just when you call and speak to the agent that, um, for example, let's take an insurance claim here. That process isn't ended when you first engage with the first either human agent or agent. Multiple things happen after that. At that point, the workflow can bring together different agents to orchestrate one end to end process.

So for example, simple customer has a car accident. That moment of emotion and difficulty generally pick up the phone. It's not something you're going to do with a chatbot or email because you're you're in a kind of emotional moment. You need instant connection. You call in, but you can speak to the ticket agent.

It's um, even I've heard today in some of the demos, even though it's a machine, they can deal with empathy and understand what they're trying to achieve and collect information about that claim. Just just to iterate. Sorry to put it in this place. When you talk about this point, you're talking about what he'll show you a demonstration of this, this case type when we talk about intent in case it worked through.

This is a type that we would never see. Self-service for first notice of loss. Someone calling in and saying that I had a collision in my automobile. Feel there's there would never be a scenario that you would particularly have for self-service perspective because of the data that needs to be collected, at least from a voice channel, that it just it would be unheard of.

And now, with the technology that's available today and what the workflow rules that we have in place, you know, Paul was able to show you that that's that's truly capable now. And then the example I've ended my claim for, and we're now past that emotional point. But I want to understand a week later, a few days later where I am in the process.

But it's not such an urgent request to pick up my pick up my phone, cell phone, and send an SMS into the into the the insurance company to find out that a genetic bot. Now it's no longer a voice, it's now a chatbot. But it can pick up that workflow. Understand who I am, do some verification, make sure it's Paul Conlin, but then give me some update on where I am in the claim, i.e.

it's a. It could be with a specific team for review Or I need to get more information. And we can even then pull in other third party agents. As a customer, I've uploaded documentation I need that needs to be verified. We don't need a human being to check the information in that document. We can. Then the workflow can call and use agent X, a third party agent GenAI agent who analyze that document, pull out the relevant information, and update that workflow.

Again, the workflow is managing the work from end to end, orchestrating many different agents, not just the agents, but also and GenAI agents in the background or even traditional robotic RPA agents as well. But it's managing that process from end to end, a much richer experience for everyone involved.

So how does it come together? So the way that we've built out, there's actually an application built into our App Studio now we call it AI studio, and you can build out the agents there. And so you'll see a listing of it in this screen. There you go. You're good. Um, and then when you go into this process, you can actually go and pick the case types that you want to leverage.

Now we're going to leverage the existing workflow rules that you have in place. The key though with this is that with, uh when you pick out these case types, we don't need a lot for training to figure out how we have. You already have your existing workflow that's put in place. Truthfully, all we really need is just a great case description.

Tell us, what is that case being used for? Give it some level of 3 to 4 sentences that would differentiate it from another one. So that when we go to do intense collection, how can I help you? Paul, what is what is the reason for your call? We're able to quickly guide them through the right workflow to be able to do that.

And then when we get to pick the case types that we put in place, we have what we call our knowledge collection. Not surprisingly, you can go in and you pick which buddies you want it to source knowledge from, right? You've built in your knowledge content. What are the sources of data that your CSRs would use if it was a human interaction? We use that same source of data to help drive that agentic process, so it knows how to properly respond.

Furthermore, you see a piece called Data Pages, right? Data pages is our type of data that we can engage with. What are the pieces they want to come in place? I'll give you an example of where this becomes extremely valuable. Imagine that I'm calling in to my bank that has issued my credit card, and I'm asking for the credit line increase my Kerim did today, right? He asked for a credit line increase. Then I go, hey, can you tell me if I'm a fraud risk? How great of a customer experience would that be for me? And then he goes, yes, Jason, we've identified that you're a moderately likely fraud risk. It's the last type of information that you'd want to share. So you have the ability to control the data sets that are available so that it's not answering, that they might ask about the Fico score or any one of those pieces.

You have control to make sure that what they're interacting with is the right type of data. Some some cases, it's going to have some level of judgment. But you want to be sure, right? You have that level of granularity of making sure that it's interacting with the right data when it has that communication.

And then finally so I have a hard time seeing this right now. I can't even see it on the screen. Even the LLM. Oh yeah. The LM of choice. So you know, Alan mentioned today, uh, choosing the right model. Now, he was mentioning it more in a statistical AI versus a large language model. But we want to give you the flexibility so that you are using the right model for the right type of interaction or type of agent that we put in place.

Paul touched base on it. We have built a framework so that these agents can communicate with other agents. We have one agent for gathering intent, another agent to servicing the case, another agent for verification. So imagine someone coming onto your website. They initiate a chat. They never logged in to their profile.

You need to make sure that it's them. I need to make sure that it's Paul on the other line. So if you're familiar with Pega Customer Service and the applications there, we have a verification process, a search and verification. We leverage that same framework from an agent perspective. And then once that agent has determined the the hey, I've got Paul, he's verified, he's authenticated.

We then hand it off to the intent. Now okay. Now we determine the intent. Now we hand it off to the agent that can do the case processing. You can do you build this these these frameworks in place, and some of them might do better with one model versus the other. Some of them might just be because you have a preference.

There are. I have seen a lot of unique conversations as we talk about generative AI that some customers are like, we only use Gemini. Well, okay, it doesn't matter if it's on our platform, it doesn't matter. We only use Gemini. Okay. No problem. Well then you can use Gemini, right? You can put in place so you have that ability of using the model of choice.

And it's a piece that will help drive through. Ready to go. Yep. All right. So now we're going to show you a demo. And you'll get to to see the power of of the voice bot. Now this is a recorded element very similar to what what Kerim did. Um this is using Blueprint. So we're going to start off with creating the case type for servicing, um, a car rental service from Vegas.

I'll feel free to chime in with some of the points that we have here if we need to articulate, but we're fast forwarding through all the Blueprint setup and then now getting into the conversation. So we're going to actually call in to the Blueprint Paul is yes, be prepared to be amazed. Understanding Scottish as.

Well. Yes. Welcome to Vegas rental Blueprint. How can I help you? And I had a little accident with the rental car. It was the Ford Ranger. Can you help out with that, please? I'm sorry to hear about the accident with the Ford Ranger. I can certainly assist you with that. Could you please provide the rental agreement number for the vehicle? Um, I have no idea what that is.

I've lost all the documentation. That's okay. We can still proceed without the rental agreement number. Could you tell me the date and time the accident occurred? It was around 2 a.m. this morning. Thank you. Could you also provide the location where the accident took place. And the Vegas Strip, Las Vegas.

Got it. Can you describe what happened during the accident and if there were any other parties involved? Um. I get dazzled by the bright lights of Vegas and get distracted. Distracted? Sorry. And drove into a traffic light. Thanks for sharing those details. Were there any other parties involved in the accident, or were you the only one affected? That's fine.

So you can see that we are able to guide through and use the workflow that was defined from Blueprint to guide through each of the steps. And we collected somewhat complicated information. Right. We're going through we're collecting each of the details that really make it challenging. And here's a perfect example of a Center-out approach.

We've used the workflow that would typically be used by a human CSR to collect that data. But now here we are immediately just making it available from point click. I mean you each of you can do this today so you can see it's real. Encourage you to build a Blueprint. If you haven't interacted with the Blueprint you never built one.

Please come down to the Innovation Hub and try and build one. And guess what? You can use that same button today that Kerim showed that said call us and then you can call on, put in your Pin number and see and you'll get that view of that. This is a true solution that is able to put in place. And as Paul mentioned earlier, you're able to deliver solutions that you can create in hours versus what you used to be many months.

And you have a much more effective and compelling solution than you ever would have had in the past. By making use of the Center-out approach. And then touching upon the Center-out approach. I'll reset it. Now we'll move on to. Then later on the same use case. As I mentioned earlier on the slides, the customer sms in to find out the state of the claim.

Again, we hear many times about omnichannel experiences. This is a true omnichannel experience. I called in and spoke to a voice box. Now I've followed up to find out the status of my claim through a chatbot. I can then upload information, a document, and then in the background. A workflow can then analyze that document, extract information from the document.

To clarify, I've got that information. So now the claims process is complete. Vegas Car Rental has all the information they need to process this claim, and hopefully it will get paid out of my claim without any real human agent being involved. Just me as a customer, engaging with an authentic voice bot and an authentic chatbot.

Is a disappointment because of that little skip on the pause. One of the things it was amazing to me is I don't know if you caught it in that sorry, Paula, I don't mean to pick on you in the Scottish accent, but you know, when he said I was when I first heard it, he sent me this video on this presentation and I'm watching it and I'm going, I have no idea what you said there.

When I was dazzled by the bright lights of Las Vegas, I had no idea I had to go back and listen to it. But it got it right. I mean, it was able to get that. And that's why I said is the machines are now more accurate than a human being. Jason could understand it, but the machine could. And I've interacted with Scottish people.

I actually led a project prior to joining Pega, where I created a model to actually detect Scottish for the the Scottish Cab company. So I've listened to it through hours of like probably thousands of hours of Scottish audio. And I mean, it's incredible that it just was able to get at that that easily for that.

So just wrapping up today some three takeaways here is that the technology is here to meet the expectations. As Jason said, coming in 2025. It's coming in 25. That's right. The voice bot will be there. So that voice is truly engaging. Voice experience powered by Pega workflows will be there for you to deliver the experience to your clients.

What was weeks? What was months? Sorry is now weeks to deliver these projects. So you're the time to value is decreased dramatically. So you can get these exceptional self-service experiences out to your customers very very quickly. So the future is here now with Pega, I think. You know, I'm surprised the room is not as full as I would expect.

I suspect that there's a a little bit of, uh, an Oversaturation of agentic so that there wasn't. But look, some of you have had some sidebar conversations with if you had this conversation with me 3 or 4 years ago and said, hey, I want to I want to automate 50% of my workflow with self-service for voice.

I would have laughed in your face. And I've been dealing in contact center technology my whole life since I was 17 years old. Right? I would have said, there's no way. Now I think 50% is too low. I think you can go much higher. I think you have a real opportunity to do a a significant amount of interaction containment that you never would have thought was possible before.

Across all of your channels. The other channels have historically been easier to deliver. Digital messaging, email, have been you've been able to do a little bit more of the interaction containment. But voice has always been historically hard. And you have a great, great, great example of it today that it's real.

It is a real item that you can put in place, and even better, you have a tool that you can use as a proof of concept that you don't even have to buy. Just go sign into Pega.com. Go create a Blueprint, call it. It's available. Go see and you can use this to advocate within your own organization of like, look, this is a reality.

This is something that we can create. This is something that we can deliver and we can produce and provide to our customers. And then I think we've. Touched upon that is get down to the Innovation Hub you can get your hands on these technologies. Mariusz. Demo booth Agentic self- service. You can go up and see this in more detail and get hands on the Agent Trainer as well.

Yeah, definitely some great AI technologies down there. If you haven't dealt with Blueprint, please go to Blueprint kiosk. Give it a shot. There is a great demonstration of our voice bot with uh, with from our self-service booth. So if you look for Agentic self self-service booth and Innovation Hub on the right hand side, you'll be able to see that, um, you know, not necessarily tied to this, but it just shows you the investments that that Pega is making into AI technologies.

If you haven't experienced the customer simulator yet, come and do that. I truly I look at it like it's almost like a carnival ride, uh, at, uh, at a user conference. I mean, every person that I've watched do this has come out of it with a smile on their face and a little bit of a shock and awe about what it's been able to produce.

So, you know, do do have that engagement. I think you'll find a lot for that. And, um, you know, try and get a little bit engaged there and see through each of those pieces. You can also see the demonstration of these technologies, uh, and our customer service desktop booth. And with that, I think we've got room for Q&A.

Any questions? Yeah. Feel free to come on up. If you've got questions. All the hard. Ones. You can go over there. We'll go side by side and. Sure. Well, thank you very much. This was great. Um, talking about predictability, I totally understand that your agent guy is based on the workflow, so you can control that.

But what if the customer goes off script, how can you control that? Yeah. So we can build in we have escalation paths and other elements that are built into that so that it's not for all the channels. So if the customer goes off the rails, it's we're not able to get the intent. We're not able to do anything for the workflow.

We we will route them to a human to try and address it if we have to. Right. That's worst case action. We didn't succeed at containment. But, you know, we're also, as you saw with the peak behind, we're keeping all the details of that interaction so you can identify why. Right. Was there a reason that led to it? Is it something that we could correct in the workflow? Like perhaps you have a workflow rule that wasn't somebody going off the rails, you just didn't have a step properly defined for this particular occurrence, and it wouldn't have only impacted your agent, it would have impacted your CSR as well.

Right? And what Jason touched upon is. That. These agents don't have access to just one workflow. You can give them access to all your workflows. And then with buddy give access to all the knowledge across your enterprise, even if it goes off piece slightly. It may be just knowledge information it needs.

It's not running a workflow, it's just simple terms and conditions that you have in your knowledge repository. Surface that up. So even when it goes off piece from what you expect the engagement to be, there's so much information that has access to, to service that customer that it may have. The challenge with knowledge is like again, the predictability, right.

You may give generate different answers depending on the case, but not understand the whole escalation path if we can stop it. Yeah. Yep. Thank you. You're welcome. Thank you. Hi. Jason and team. It's a great overview of the Platform feature. My question is more of the orchestration within the agents.

Like a kind of a saga pattern. What you call, um, say, say I'm in the conversational bot with the CSR, talking to the bot in the prompt engineering, considering that as a master agent and now I have a repetitive, repetitive task for system driven tasks that I need to scale up to split, that can go into systemic autonomy or autonomous task that can be automated through agents.

So I need to create a master slave agents, like kind of talking to each other and clone that agents and scale their agents. So what are the opportunities or the platform capabilities that like roadmap, do you have any roadmap from now from here to how do I scale and go to that pattern? So I would like to say that the roadmap is here. So we have a concept kind of along the lines of what you talked about, which we call agent chaining. So as one agent solves one item and it makes a determination that it needs another agent to do this. So if we're thinking about we've been communicating almost from a like a conversational perspective.

Right. Because that's really where a lot of the focus and it's helped people, helps people visualize. But you're talking about more of like a back office type scenario, right? I've got chunks of work that typically would require a human to do these steps and follow these items through. It can do that.

And then as it advances to the next step, it can pass that contextual data for what it just processed and what it just solved off to the next agent. So that agent chaining is natively available in the platform to handle that. And from what we presented here, that same element of the chaining is relevant from a conversational perspective, right.

So when you look at it from a customer service perspective, it's there from a platform perspective, it's there as well. You can use that chaining to solve more complex workflow. Handing it off. And quite honestly, we don't want you in prompt engineering like our goal. When Ellen mentioned that you noticed when I showed you those UIs, did you see a prompt? No.

You saw a model, right? Like we we are going to you know, our expectation on this release is that could you tweak the prompt to make adjustments? Absolutely. We're not going to prevent you from doing that. But we did our job right. You didn't need to, right? It was able to work through the workflow and our rules, and it was able to be successful.

Thank you. You're welcome. So my question might be difficult to articulate, but we've optimized the mechanics of development testing deployments so things can be done very quickly. But in highly regulated environments, compliance teams, they're the ones who that's the hoop you got to get through. Are there steps you're taking to help explain, produce documentation, or show or teach the compliance teams on this AI model? Here's how it's going to treat customers fairly.

Here's how it's here's what the change looks like. So that because otherwise it doesn't matter how fast you can develop something if they can't understand it, especially if it's AI versus explicit business rules, it's going to take long. It's going to take maybe longer. So so the answer to that, that's a little bit of a challenging item, right.

Because how do you get past the I fear. So here are the things that we did to help with that. We have a transaction history. You saw the peek behind. So you talk about text transcript that's available from a voice perspective. We have the capability to record those interactions as well. So your compliance team that doesn't have the comfort level that they're where you're giving the same experience that Paul as a human CSR would.

You can give them proof and evidence that absolutely we can. And here it is. Here's, you know, a dozen three minute calls. Please go spend 40 minutes and listen to it. Right. And then they can gain that level of confidence and put in place. You know, the other element of it is, is that when you have those workflow rules and you put in place, we can leverage the same compliance structures you have for messaging that needs to be put in place as well, right? So you know where you have, uh, strict dialogs, strict terms and conditions, those types of items, we can leverage those items that are built into your workflows, and read those off to make sure that your AI agent is still compliant and meeting the demands of your business officers compliance officers in your team. I would just make the comment that typically we try to bring them in early to avoid they're not going to sign, you're not going to get they're not going to look for call logs because we we aren't at that point yet. Yeah. They need comfort that what we're going to develop is going to be, like I said, not a black box.

Yeah, yeah. And I mean, that's a difficult piece. I mean, when you develop it in. But the good news is your, your timeline to being able to have those conversations is much quicker, right? I mean, typically you would have been let's talk through your workflow. You might have went in and had a conversation with your compliance officers and said, hey, we're going to build this new self-service solution for our voice.

It's going to do these things. What do you think? And they would have been like, okay, well, you got to do these seven steps that we would have expected the human to do. And what you would have done is you would have went back prior to today. Right? Prior to the world of today with technology and four months later you would have come back and said, okay, we've got it.

What do you think now? I mean, you've built it for your CSRs. You've basically you can prove to them that it's using the same workflow, and then you're able to to build in that confidence to help that conversation much quicker. Appreciate it. You're welcome. Any other questions in the group? Great demo by the way.

Um, this is from Northwestern Mutual. Uh, we are going through some Voice AI demos right now. A curious question around customer service, and some of the guided authentication flows are built in Pega customer service. Have you thought about a Agentic AI just to handle the authentication caller authentication? In theory, you could do that.

So great question. So this question is hey, let's take the CSR out of the process of doing the authentication and verification when it comes in let's say the the Caller but you know visitor your interaction goes into your existing solution. I'm assuming you probably have an IVR that asks them questions, but they're just hitting 00000 and then they get to your CSR.

And now you got to ask them all the questions that you had the answer to. That is yes. Yes you could. You could have when your IVR or when that inbound call comes to you or, you know, start with our agentic process and then the next step be okay. Now that we've verified we're going to escalate to human agent, we're going to escalate it off.

We're going to hand it across so they don't have to do that. And it comes in pre verified for it. It is possible to do. Absolutely. Thank you. You're welcome okay. Anything else. All right. Well we made it with just a few seconds time. We truly appreciate your attendance. Hopefully you found it valuable.

Thanks, Joe. Come see us in the Innovation Hub, please..